Protecting our FLOSS commons from LLMs (blog.codeberg.org)

Codeberg e. V. recently held a vote that resulted in two motions aimed at curtailing the influence of Large Language Models (LLMs) within their free and open source software ecosystem. The first motion is a declaration of intent not to use users' data for training LLMs, emphasizing a commitment to user privacy and the integrity of their projects. The second, more contentious motion modifies the Terms of Use to restrict 'vibe-coded' projects, which are often characterized by low community engagement and high resource consumption. This decision highlights Codeberg’s concerns regarding the sustainability and collaborative spirit of the software development community, as reliance on LLMs creates one-off projects that lack ongoing maintenance and community involvement. The implications of these motions are significant for the AI and ML landscape, particularly concerning ethical practices around data usage and software development. By prohibiting the use of code and data to train LLMs, Codeberg positions itself against the trend of LLM-driven software creation, which can dilute collaboration and innovation within the free software community. Additionally, the rising costs attributed to LLM training and the environmental impact of data centers supporting them have prompted calls for more sustainable and responsible practices. Codeberg aims to protect its platform as a space for concentrated human collaboration, resisting the commodification and isolation that LLMs can bring to software development.